Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
1
fork

Configure Feed

Select the types of activity you want to include in your feed.

at v2.6.13-rc2 286 lines 11 kB view raw
1 2Device Drivers 3 4struct device_driver { 5 char * name; 6 struct bus_type * bus; 7 8 struct completion unloaded; 9 struct kobject kobj; 10 list_t devices; 11 12 struct module *owner; 13 14 int (*probe) (struct device * dev); 15 int (*remove) (struct device * dev); 16 17 int (*suspend) (struct device * dev, pm_message_t state, u32 level); 18 int (*resume) (struct device * dev, u32 level); 19}; 20 21 22 23Allocation 24~~~~~~~~~~ 25 26Device drivers are statically allocated structures. Though there may 27be multiple devices in a system that a driver supports, struct 28device_driver represents the driver as a whole (not a particular 29device instance). 30 31Initialization 32~~~~~~~~~~~~~~ 33 34The driver must initialize at least the name and bus fields. It should 35also initialize the devclass field (when it arrives), so it may obtain 36the proper linkage internally. It should also initialize as many of 37the callbacks as possible, though each is optional. 38 39Declaration 40~~~~~~~~~~~ 41 42As stated above, struct device_driver objects are statically 43allocated. Below is an example declaration of the eepro100 44driver. This declaration is hypothetical only; it relies on the driver 45being converted completely to the new model. 46 47static struct device_driver eepro100_driver = { 48 .name = "eepro100", 49 .bus = &pci_bus_type, 50 51 .probe = eepro100_probe, 52 .remove = eepro100_remove, 53 .suspend = eepro100_suspend, 54 .resume = eepro100_resume, 55}; 56 57Most drivers will not be able to be converted completely to the new 58model because the bus they belong to has a bus-specific structure with 59bus-specific fields that cannot be generalized. 60 61The most common example of this are device ID structures. A driver 62typically defines an array of device IDs that it supports. The format 63of these structures and the semantics for comparing device IDs are 64completely bus-specific. Defining them as bus-specific entities would 65sacrifice type-safety, so we keep bus-specific structures around. 66 67Bus-specific drivers should include a generic struct device_driver in 68the definition of the bus-specific driver. Like this: 69 70struct pci_driver { 71 const struct pci_device_id *id_table; 72 struct device_driver driver; 73}; 74 75A definition that included bus-specific fields would look like 76(using the eepro100 driver again): 77 78static struct pci_driver eepro100_driver = { 79 .id_table = eepro100_pci_tbl, 80 .driver = { 81 .name = "eepro100", 82 .bus = &pci_bus_type, 83 .probe = eepro100_probe, 84 .remove = eepro100_remove, 85 .suspend = eepro100_suspend, 86 .resume = eepro100_resume, 87 }, 88}; 89 90Some may find the syntax of embedded struct initialization awkward or 91even a bit ugly. So far, it's the best way we've found to do what we want... 92 93Registration 94~~~~~~~~~~~~ 95 96int driver_register(struct device_driver * drv); 97 98The driver registers the structure on startup. For drivers that have 99no bus-specific fields (i.e. don't have a bus-specific driver 100structure), they would use driver_register and pass a pointer to their 101struct device_driver object. 102 103Most drivers, however, will have a bus-specific structure and will 104need to register with the bus using something like pci_driver_register. 105 106It is important that drivers register their driver structure as early as 107possible. Registration with the core initializes several fields in the 108struct device_driver object, including the reference count and the 109lock. These fields are assumed to be valid at all times and may be 110used by the device model core or the bus driver. 111 112 113Transition Bus Drivers 114~~~~~~~~~~~~~~~~~~~~~~ 115 116By defining wrapper functions, the transition to the new model can be 117made easier. Drivers can ignore the generic structure altogether and 118let the bus wrapper fill in the fields. For the callbacks, the bus can 119define generic callbacks that forward the call to the bus-specific 120callbacks of the drivers. 121 122This solution is intended to be only temporary. In order to get class 123information in the driver, the drivers must be modified anyway. Since 124converting drivers to the new model should reduce some infrastructural 125complexity and code size, it is recommended that they are converted as 126class information is added. 127 128Access 129~~~~~~ 130 131Once the object has been registered, it may access the common fields of 132the object, like the lock and the list of devices. 133 134int driver_for_each_dev(struct device_driver * drv, void * data, 135 int (*callback)(struct device * dev, void * data)); 136 137The devices field is a list of all the devices that have been bound to 138the driver. The LDM core provides a helper function to operate on all 139the devices a driver controls. This helper locks the driver on each 140node access, and does proper reference counting on each device as it 141accesses it. 142 143 144sysfs 145~~~~~ 146 147When a driver is registered, a sysfs directory is created in its 148bus's directory. In this directory, the driver can export an interface 149to userspace to control operation of the driver on a global basis; 150e.g. toggling debugging output in the driver. 151 152A future feature of this directory will be a 'devices' directory. This 153directory will contain symlinks to the directories of devices it 154supports. 155 156 157 158Callbacks 159~~~~~~~~~ 160 161 int (*probe) (struct device * dev); 162 163The probe() entry is called in task context, with the bus's rwsem locked 164and the driver partially bound to the device. Drivers commonly use 165container_of() to convert "dev" to a bus-specific type, both in probe() 166and other routines. That type often provides device resource data, such 167as pci_dev.resource[] or platform_device.resources, which is used in 168addition to dev->platform_data to initialize the driver. 169 170This callback holds the driver-specific logic to bind the driver to a 171given device. That includes verifying that the device is present, that 172it's a version the driver can handle, that driver data structures can 173be allocated and initialized, and that any hardware can be initialized. 174Drivers often store a pointer to their state with dev_set_drvdata(). 175When the driver has successfully bound itself to that device, then probe() 176returns zero and the driver model code will finish its part of binding 177the driver to that device. 178 179A driver's probe() may return a negative errno value to indicate that 180the driver did not bind to this device, in which case it should have 181released all reasources it allocated. 182 183 int (*remove) (struct device * dev); 184 185remove is called to unbind a driver from a device. This may be 186called if a device is physically removed from the system, if the 187driver module is being unloaded, during a reboot sequence, or 188in other cases. 189 190It is up to the driver to determine if the device is present or 191not. It should free any resources allocated specifically for the 192device; i.e. anything in the device's driver_data field. 193 194If the device is still present, it should quiesce the device and place 195it into a supported low-power state. 196 197 int (*suspend) (struct device * dev, pm_message_t state, u32 level); 198 199suspend is called to put the device in a low power state. There are 200several stages to successfully suspending a device, which is denoted in 201the @level parameter. Breaking the suspend transition into several 202stages affords the platform flexibility in performing device power 203management based on the requirements of the system and the 204user-defined policy. 205 206SUSPEND_NOTIFY notifies the device that a suspend transition is about 207to happen. This happens on system power state transitions to verify 208that all devices can successfully suspend. 209 210A driver may choose to fail on this call, which should cause the 211entire suspend transition to fail. A driver should fail only if it 212knows that the device will not be able to be resumed properly when the 213system wakes up again. It could also fail if it somehow determines it 214is in the middle of an operation too important to stop. 215 216SUSPEND_DISABLE tells the device to stop I/O transactions. When it 217stops transactions, or what it should do with unfinished transactions 218is a policy of the driver. After this call, the driver should not 219accept any other I/O requests. 220 221SUSPEND_SAVE_STATE tells the device to save the context of the 222hardware. This includes any bus-specific hardware state and 223device-specific hardware state. A pointer to this saved state can be 224stored in the device's saved_state field. 225 226SUSPEND_POWER_DOWN tells the driver to place the device in the low 227power state requested. 228 229Whether suspend is called with a given level is a policy of the 230platform. Some levels may be omitted; drivers must not assume the 231reception of any level. However, all levels must be called in the 232order above; i.e. notification will always come before disabling; 233disabling the device will come before suspending the device. 234 235All calls are made with interrupts enabled, except for the 236SUSPEND_POWER_DOWN level. 237 238 int (*resume) (struct device * dev, u32 level); 239 240Resume is used to bring a device back from a low power state. Like the 241suspend transition, it happens in several stages. 242 243RESUME_POWER_ON tells the driver to set the power state to the state 244before the suspend call (The device could have already been in a low 245power state before the suspend call to put in a lower power state). 246 247RESUME_RESTORE_STATE tells the driver to restore the state saved by 248the SUSPEND_SAVE_STATE suspend call. 249 250RESUME_ENABLE tells the driver to start accepting I/O transactions 251again. Depending on driver policy, the device may already have pending 252I/O requests. 253 254RESUME_POWER_ON is called with interrupts disabled. The other resume 255levels are called with interrupts enabled. 256 257As with the various suspend stages, the driver must not assume that 258any other resume calls have been or will be made. Each call should be 259self-contained and not dependent on any external state. 260 261 262Attributes 263~~~~~~~~~~ 264struct driver_attribute { 265 struct attribute attr; 266 ssize_t (*show)(struct device_driver *, char * buf, size_t count, loff_t off); 267 ssize_t (*store)(struct device_driver *, const char * buf, size_t count, loff_t off); 268}; 269 270Device drivers can export attributes via their sysfs directories. 271Drivers can declare attributes using a DRIVER_ATTR macro that works 272identically to the DEVICE_ATTR macro. 273 274Example: 275 276DRIVER_ATTR(debug,0644,show_debug,store_debug); 277 278This is equivalent to declaring: 279 280struct driver_attribute driver_attr_debug; 281 282This can then be used to add and remove the attribute from the 283driver's directory using: 284 285int driver_create_file(struct device_driver *, struct driver_attribute *); 286void driver_remove_file(struct device_driver *, struct driver_attribute *);